measuring visual quality of street space and its temporal ...€¦ · “physical vqoss”)...

Contents lists available at ScienceDirect

Landscape and Urban Planning

journal homepage: www.elsevier.com/locate/landurbplan

Research Paper

Measuring visual quality of street space and its temporal variation:Methodology and its application in the Hutong area in BeijingJingxian Tang, Ying Long⁎

School of Architecture and Hang Lung Center for Real Estate, Tsinghua University, Beijing, China

A R T I C L E I N F O

Keywords:Public spaceVisual qualityPerceived qualityStreet view pictureHutong

A B S T R A C T

Although it is widely known that quality of street space plays a vital role in promoting urban vibrancy, there isstill no consensus on how to quantitatively measure it for a large scale. Recent emerging dataset Street ViewPicture has revealed the possibility to overcome the previous limit, thus bringing forward a research paradigmshift. Taking this advantage, this paper explores a new approach for visual quality evaluation and variationidentification of street space for a large area. Hutongs, which typically represent for historical street space inBeijing, are selected for empirical study. In the experimental part, we capture multi-years Tencent Street ViewPicture covering all the Hutongs, and conduct both physical and perceived visual quality evaluation. Thephysical visual quality of street space is achieved automatically by combining 3-dimensional composition cal-culation of greenery, openness, enclosure using machine-learning segmentation method SegNet, and 2-dimen-sional analysis of street wall continuity and cross-sectional proportion; perceived visual quality of street space isevaluated by stay willingness scoring from five aspects. The variation of quality is evaluated based on theidentified physical space variations. The result indicates that visual quality of Hutongs are not satisfied, whilesome regeneration projects in the historical protection block is better. Most Hutongs are in shortage of visualgreen, relative more continuous but with low cross-sectional ratio. Hutongs near main road witness an increasingtrend of motorization. The difference between physical and perceived quality indicates the feasibility and lim-itation of the auto-calculation method. In the most recent 3–4 years, less than 2.5% Hutongs are improved,which are mainly slow beautification.

1. Introduction

Streets, together with parks, plazas and squares, have been widelydeemed as public space, which is an important arena for urban vitality.Streets, as a vital component of the built environment, are highlightedfor their values across various dimensions, such as traffic, aesthetics,public health and neighborhood interaction (Jiang, Chang, & Sullivan,2014; Mounir, Attia, & Tayel, 2016; Middleton, 2016). Good streetenvironments could enhance the frequency of outdoor activities, thusaffecting the users’ behavior. Fuelled by the increasing emphasis onQuality of Street Space (termed “QoSS” in this paper) and lively builtenvironment, designing human-oriented street space has become one ofthe crucial environmental improvement strategies for competitive ci-ties. Various initiatives have been proposed for streets, such as BetterStreets in London, Green Streets in the city of Portland, Cycle Strategy inCopenhagen, and Street Design Guidelines in Shanghai. With this back-ground, recent years have seen the emergence of several street en-vironment evaluation platforms like Walk Score, State of Place Index,

Ratemystreet, and Walkability Asia. The widespread street policies andplatforms indicate a global consensus on highlighting QoSS in additionto extensive discussion on the traffic function of streets.

This call for the renewed livelihood of streets is not the first attempthistorically. Efforts trace back to the 1960s when Jacobs (1961) usedsocial descriptions to express the pursuit of neighborhood life. Otherscholars also interpreted it into design vocabularies to define the con-cept of good street space (Lynch, 1960; Ewing & Clemente, 2013).However, due to the time-consuming nature of field surveys, nearly allurban designers and researchers at the time were restricted to small-scale empirical studies, finding connections between physical appear-ance and social attributes and deepening their logical reasoningthrough induction and deduction, instead of principle exploration basedon large-scale investigation. Automatic quantitative measurement ofspatial QoSS for large areas has long faced difficulties, and problemscompound with temporal variation. This situation prevents plannersand designers from scientifically proposing a rational blueprint forstreets.

https://doi.org/10.1016/j.landurbplan.2018.09.015Received 14 July 2017; Received in revised form 29 August 2018; Accepted 16 September 2018

⁎ Corresponding author.E-mail address: [email protected] (Y. Long).

Landscape and Urban Planning 191 (2019) 103436

Available online 05 October 20180169-2046/ © 2018 Elsevier B.V. All rights reserved.

T

http://www.sciencedirect.com/science/journal/01692046

https://www.elsevier.com/locate/landurbplan

https://doi.org/10.1016/j.landurbplan.2018.09.015


mailto:[email protected]


http://crossmark.crossref.org/dialog/?doi=10.1016/j.landurbplan.2018.09.015&domain=pdf

The emerging dataset Street View Picture (SVP) and cutting-edgetechniques like deep learning algorithms for processing images shedlight on measuring the visual aspect of QoSS and its variation over alarge geographical area. This study builds a methodology to measurethe visual Quality of Street Space (vQoSS) as well as its temporalvariations; the approach is then applied to historical streets in Beijing(Hutongs). The research questions are as follows: (1) What is the bestway to measure the physical visual quality of street space (termed“physical vQoSS”) automatically on a large scale? (2) What is theoverall perceived visual quality of street space (termed “perceivedvQoSS”) in the study area, and to what extent can the physical vQoSSrepresent the perceived one? (3) What is the best method for detectingthe physical changes of street components so as to measure the vQoSSvariation over time?

This paper reviews the related definitions, theories, and approachesregarding quantitative measurement of vQoSS and its variation inSection 2. This is followed by an introduction of the study area and datacollection, as well as details about the research methods in Sections 3and 4, respectively. In Section 5, an empirical study of Hutongs hasbeen conducted to test and explore the feasibility and limitations of themethods. The paper also reflects upon the status quo and the changingtrends of temporary design for the historical area in Beijing. The eva-luation result of Hutongs and the visual quality changes in recent yearsare examined. The last section gives the concluding remarks, and pro-poses potential applications and limitations of the study.

2. Literature review

2.1. Defining “visual quality of street space” (vQoSS)

In this paper, the concept of “quality” is used to address multi-as-pect attributes of street space. Quality generally refers to a degree ofexcellence in something, and it is applied to the concept of street spacesas it relates to their environmental conditions and level of service. Fruin(1971) emphasizes the synergy of safety, security, convenience, con-tinuity, comfort, and attractiveness of streets in general; Slater (1985)proposed the concept of comfort and classified street comfort into threecategories, namely physical, psychological, and physiological. Physio-logical comfort is perceived through different sensors such as color,light, smell, sound and thermal comfort, which are based on the per-ception of the physical compositions of street space. To be more spe-cific, a street equipped with appropriate form and perceptual designmay gain increased physical function and emotional appeal, thushaving restorative potential as mentioned previously by Kaplan (Kaplan& Kaplan, 1989; Kaplan, 1995). In addition, the social meaning orsymbolic image of a space can be attributed to its perceived feeling ingeneral (Lynch, 1960; Thompson, 2002; Fyfe, 2006). The overallquality generally invokes a combination of visual and non-visual feel-ings, influenced by weather, climate, pollution, human activities, aswell as the context attached to certain history (Nasar, 1989; Cassidy,1997; Johansson, Sternudd, & Ferreira, 2015).

During the last ten years, a considerable number of studies havemeasured the multisensory qualities of street space, including theirauditory quality (Pheasant, Horoshenkov, Watts, & Barret, 2008;Easteal, Bannister, & Kang, 2014; Aletta, Lepore, Kostara-Konstantinou,Kang, & Astolfi, 2016; Aiello, Schifanella, Quercia, & Aletta, 2016), thesmellscape (Henshaw, 2014), and the visual streetscape (Li, Ratti, &Seiferling, 2017; Harvey, Aultman-Hall, Hurley, & Troy, 2015; Naik,Kominers, Raskar, Glaeser, & Hidalgo, 2015; Salesses, Schechtner, &Hidalgo, 2013). In this paper, the focus is the development of metho-dology for measuring the visual streetscape. vQoSS is the comprehen-sive environmental condition and users’ physical reflection of the staticimage of the space, including its color, texture, scale, and styles. On theone hand, vQoSS can be inferred from users’ willingness to remain inthe space, while on the other, it objectively relies on physical compo-nents of streetscape, namely tree canopies (greenery), enclosures,

openness, connectivity, street wall continuity, density, accessibility,cross-sectional proportion, scale, tidiness and so on (Harvey, Aultman-Hall, Hurley, & Troy, 2015; Alexander et al., 1977; Saelens, Sallis, &Frank, 2003).

Classic academic works have repeatedly proven that physical spaceand street form are the cornerstones of street activities ( Lynch, 1984;Gehl & Svarre, 2013; Gehl, 2013). From the perspective of environ-mental psychology, they play a fundamental role in social interaction,while good form brings better perceived visual feeling. In sum,greenery, openness, enclosures, street wall continuity, and cross-sec-tional proportion, are some of the basic and over-discussed morpholo-gical features among all the variables that contribute to the vQoSS.

Enclosure forms space; it represents the extent of human-scale; awell enclosed street tends to produce an impression of higher securityfor users, thus providing more opportunity for physical activities(Arnold, 1980; Alexander et al., 1977; Owens, 1993; Rapoport, 2016;Ewing & Handy, 2009), while a wide set-back structure always gen-erates a feeling of emptiness and inactivity (Dover & Massengale, 2013;Montgomery, 2013). Therefore, enclosure is highlighted as the primaryfeature in designs that amplify street life and it is widely used as ameasurement of comfort. Continuity could be defined as the propor-tion of street edge intersecting with buildings (Harvey, 2014; Heathet al., 2006); a complete and continuous building façade generates astreetscape with order and vitality within the provided enclosure(Ewing & Handy, 2009; Gehl & Svarre, 2013). With commercial fron-tage, a continuous street will attract more pedestrians in its closed-offatmosphere (Gehl, 2013). The contribution of greenery to vQoSS andperceived safety has been repeatedly documented (Li, Zhang, & Li,2015, Li, Zhang, Li, Ricard, et al., 2015). Jiang et al. (2015) proved thattree cover density influences peoples’ stress, and a tree canopy aug-ments the enclosed feeling.

Cross-sectional proportion affects pedestrians’ perception of scale,influencing their willingness to remain in a space (Dover & Massengale,2013; Ewing & Handy, 2009). A wider street always implies highermotorization with more traffic, and potential danger, pollution, noiseand car accidents. By contrast, a relatively narrow street encouragespublic leisure (Gehl & Svarre, 2013; Gehl, 2013). Harvey et al. (2015)verify that the skeletal proportions of streetscapes across New York Cityaccounted for approximately 42% of the variability in perceived safety,and both enclosing buildings and tree canopies had a positive impact onvQoSS. Last but not least, openness, the degree of sky visibility, de-termines the amount of perceived lightness, having an impact on visualperception and pleasantness (Bruce, Green, & Georgeson, 2003; Li et al.,2017), and microclimate (Lin, Tsai, Hwang, & Matzarakis, 2012;Grimmond, Potter, Zutter, & Souch, 2001; Unger, 2008). Besides thesefive aspects, there are many indicators that reflect the physical qualityof the street. However, in this paper, a synthesis method is demon-strated against the currently measurable ones. Other variables could beadded into the framework gradually in further empirical and theoreticalstudies.

2.2. Key methods for measuring vQoSS

Academic discussions and applications of quantitative measurementof vQoSS have revolved around five categories: (1) subjective percep-tion assessment; (2) systematic observation and rating of streetscape;(3) physiological monitoring and analysis; (4) laboratory experimentsand analysis; and (5) computer-assisted auditing and evaluation. Inaddition, some discussions integrate two of the above methods for onelocation. These are generally addressed in Table 1.

2.2.1. Subjective perception assessmentSubjective perception assessment is the method of evaluating re-

spondents’ perceived emotions to the selected built environmentthrough different kinds of interviews or in-person questionnaires. Theresults could help to identify important factors in the built environment

J. Tang and Y. Long Landscape and Urban Planning 191 (2019) 103436

2

that affect physical activities (Brownson, Hoehner, Day, Forsyth, &Sallis, 2009). For example, Appleyard and Lintell (1972) carried outfield interviews on three similar streets with different traffic levels inSan Francisco to find out how traffic conditions affected the livabilityand quality of the street environment. The measurement includes fiveindicators, namely perceived livability, levels of social interaction,territorial extent, environmental awareness and safety for street quality.

2.2.2. Systematic observation and rating of streetscapeSystematic observation and rating is the procedure to obtain a

complete understanding of the physical features of streetscape. Thedata source for this method includes survey images, video clips(taping), and SVPs. Evaluation indexes with different indicators havebeen constructed to meet a baseline requirement to measure theotherwise immeasurable (Ewing & Clemente, 2013; Ye & van Nes,

2014). These indicators convert the blurry conception of vQoSS intounderstandable vocabularies related to urban design, facilitating de-signers’ pre-assessment and shaping of design ideas. For example,Ewing and Clemente (2013) construct a systematic evaluation indexincluding Enclosure, Human scale, Transparency, Tidiness and Image-ability to represent street quality. Tang and Long (2017) decompose thevQoSS according to the above five dimensions and attach a subjectivescore from designers. Human observers allow audits to account forvariables that are subjective and nuanced, but sometimes impractical toevaluate from existing spatial datasets (Harvey, 2014). It is a reliableand preferable method once we overcome the limitations in audit scaleand raise efficiency with the help of new technology.

2.2.3. Physiological monitoring and analysisSome researchers conduct technologically advanced evaluation with

Table 1Classification of methods for measurement of vQoSS.

Category Sub-Category Sample Research method

A. Subjective perceptionassessment

Interview McGinn, Evenson, Herring,Huston, and Rodriguez(2007)

Perception of the built environment gathered from responses to 1270telephone surveys.

In-person questionnaires Sallis, Johnson, Calfas,Caparosa, and Nichols(1997)

43 in-person questionnaires were used to assess environmental variables athomes, in neighborhoods, or on frequently traveled routes.

B. Systematic observation andrating of streetscape;

Based on site survey Gehl and Gemzøe (1996) Investigation on Public Life and Public Space (PLPS), measuring quality ofpublic space, quality of the facade along the street, pedestrian volume, andactivities.

Based on video clip or videotaping Ewing and Clemente (2013) Construction of systematic evaluation index, including five aspects torepresent the quality of the street, and the video recording of interviewees’subjective ratings of the street.

Based on large -scale collected SVP Tang et al. (2016) Subjective rating of the user’s willingness to remain in the street spaceusing multi-year SVPs, to measure the comprehensive quality andtemporal variation.

Curtis, Curtis, Mapes, Szell,and Cinderich (2013)

Using Google Street View GSV as a systematic observation tool for builtenvironment audits.

Rundle, Bader, Richards,Neckerman, and Teitler(2011)

Evaluation of the aesthetics and other physical aspects of theneighborhood with SVPs.

C. Physiological monitoringand analysis

Electrophysiological evaluation bysensors

Jiang et al. (2014) Measurement of the participants’ stress in different eye-level tree coverdensities by measurement of skin conductance and salivary cortisol levels.

Observation of users with otherwearable devices

Aspinall et al. (2013) Observation of the differences of Electroencephalograms (EEG)characteristic of participants’ response in different walking environments,with the help of Emotiv EPOC recorders.

Eye-tracking Metrics Dupont, Antrop, and VanEetvelde (2014)

Using Eye-tracking Metrics (ETM) to investigate how people perceive andobserve landscape images (not street quality, but similar) in the lab.

D. Laboratory experiments andanalysis

Virtual reality Luigi et al. (2015) Using Immersive Virtual Reality (IVR) as a tool to stimulate the differenthuman senses and create the experience of the sense of “being there”.

Augmented reality (AR) Carozza et al. (2014) A novel AR system based on monocular vision is built to calculateocclusions of the built environment on augmenting virtual objects.

E. Computer-assisted auditingand evaluation

GIS and remote sensing analysisbased on 2D data

Kendal, Williams, andWilliams (2012)

Using aerial photos with a 10 cm resolution as the data source andconducting GIS analysis of tree cover in gardens, parks, and streetscapes.

Harvey (2014) Measuring continuity and cross-sectional proportion of streetscape withGIS.

Visual analysis or auditing tool ofSVP

Bader et al. (2015) Using GSV to conduct virtual audits of neighborhood environments withComputer Assisted Neighborhood Visual Assessment System (CANVAS).

Li, Zhang, Li, Ricard, et al.(2015)

Green vegetation extraction from GSV and calculation of the green viewindex of the streetscape.

Machine-learning-assisted methodfor large-scale SVP

Salesses et al. (2013) Evaluation of street quality with regard to the aspects of perception ofsafety, class and uniqueness based on a dataset created by Place Pulse.

Naik et al. (2014) Using ‘Streetscore’, a scene-understanding algorithm to predict theperceived safety of a streetscape, using training data from an online surveywith contributions from more than 7000 participants.

Harvey et al. (2015) Identification of a set of “streetscape skeleton” design variables, andmeasurement of these variables’ effect on a large and diverse sample ofstreetscapes, examining their relationship to perceived safety scorescreated by Place Pulse.

Tang and Long (2017) Evaluation of quality with both subjective rates from five aspects createdby Ewing and Clemente (2013), and objective component analysis usingimage segmentation by SegNet.

Shen et al. (2018) Integration of visualization techniques with machine learning models tofacilitate the detection of street view patterns with StreetVizor.


3

the help of biosensors to record the physiological response of pedestriangroups in different street spaces, for example using salivary cortisolmeasurement, Eye-tracking Metrics (ETM), Electroencephalograms(EEG), nuclear magnetic resonance (NMR), and hormone tests. Thesetechnologies provide a new perspective for street quality measurement.For example, Aspinall, Mavros, Coyne, and Roe (2013) investigate theuse of mobile EEG to record the emotional experience of a group ofwalkers in three types of street space. Jiang (2015) measured partici-pants’ stress level changes in neighborhood street scenes through hor-monal (salivary cortisol), physiological (skin conductance) and psy-chological (self-reported stress) variables. These devices enable theobservation of specific physiological parameters of users, reflecting thedifferences between physical street environments.

2.2.4. Laboratory experiments and analysisSeveral studies have emphasized the possibility of Virtual or aug-

mented reality (VR and AR) as visual stimuli for subjective perceptionresearch in the laboratory (Wang, Kim, Love, & Kang, 2013). Combinedwith radio processing and 3D modeling, a simulated environment withcontrollable variables could be set in VR, enabling the researchers tobuild experiments to evaluate perception, which could be integrated inthe urban design process (Luigi, Massimiliano, Aniello, Gennaro, &Virginia, 2015). AR and VR provide advantages for enhancing state-of-the-art visual scenes through proposed designs within the real en-vironment (Carozza, Tingdahl, Bosché, & Gool, 2014).

2.2.5. Computer-assisted auditing and evaluationComputer-assisted auditing and evaluation becomes increasingly

popular due to efficiency. With GIS and remote sensing, aerial photoswith high resolution are used for top view analysis. And the emergenceof the Street View Picture dataset has brought more possibilities. SVPscreate a continuous 360° image of a streetscape (Li, Zhang, & Li, 2015,Li, Zhang, Li, Ricard et al., 2015), allowing users to easily imagine thereal vQoSS scene. It presents all surrounding information equivalent toan on-the-spot investigation on a larger scale at one time. In addition,companies such as Tencent even have a multi-year picture recordingfunction to record a time series of variance in street space. In recentyears, increasing numbers of quantitative methodologies and empiricalinvestigation studies have been explored to use SVP data in existingmethods. Harvey (2014) identifies four streetscape skeleton types ofthree American cities with the help of SVPs and GIS. Combined withmachine-learning, MIT Media Lab launched a project called Place Pulse,aiming to quantitatively identify which areas of a city are perceived aswealthy, modern, safe, lively, active, central, adaptable or familyfriendly, through the online website Place Pulse (http://pulse.media.mit.edu) and large-scale SVPs. Further studies by Naik, Philipoom,Raskar, and Hidalgo (2014), Salesses, Schechtner, and Hidalgo (2013),Harvey, Aultman-Hall, Hurley, and Troy (2015), and Li, Zhang, and Li(2015) were based on the dataset of Place Pulse. The latest excitingmethod from SegNet presents a deep learning framework for semanticsegmentation of SVPs through which 12 categories of objective ele-ments are recognized (Kendall et al., 2015). Together with other similarmethods, this breakthrough increases the potential for tremendousadvances in the field of urban design, especially in studies of the builtenvironment. Tang and Long (2017) firstly apply image segmentationin their comparative study of the street quality in central Beijing andShanghai, illustrating its feasibility.

2.3. Key factors for identifying variations of vQoSS

The automatic quantification of spatial variation is not easy toachieve. Social scientists interested in gentrification apply communitylevel statistical data to discover the interaction between spatial changeand the social attributes instead of using real spatial appearance vari-ables. Computational science brings a recent breakthrough. Both per-ceived and physical variation detection are shown in Table 2. Naik et al.

(2015) firstly use the Urban Change Coefficient, calculating the dynamicperceived safety variation over a period of time derived from an algo-rithm checking SVPs. Using deep Convolutional Neural Networks andthe applied algorithm, streetscape level variations in pictures and vi-deos can be identified. Physical variation identification through scoringis also more accurate, and avoids the errors generated by view angles.Tang, Ma, Zhai, and Long (2016) divide the physical street space intofour parts according to sections and identify the temporal variation,building an evaluation system on physical change and its effectiveness,which indicates visual quality differentiations from year to year. Thismethod is relatively reliable and easy to carry out, and is therefore usedin identification of Hutong variation to reflect the change of vQoSS inthis study.

Based on the above understandings, this research proposes a com-prehensive approach for evaluation and variation identification ofvQoSS with general applicability, not only with the traditional sub-jective scoring by designers but also with the automated recognition ofthe physical visual quality to reduce subjective bias.

3. Methodology

3.1. Study area and data collection

3.1.1. Study areaOur research is conducted within the Second Ring Road of Beijing

(with an area of 62 km2), also called the Old City, which dates back toYuan Dynasty. Fig. 1 presents the road map of the area. In the Ming andQing Dynasty, the Old City was the outer boundary of Beijing, mainlyconsisting of main roads (Da jie), secondary roads (Xiao jie), alleys andthe Chinese traditional courtyard (siheyuan). According to the RecordDraft on Blocks and Streets in Beijing by Zhu Yixin in the Qing Dynasty,there were about 3500 alleys and 978 of them were directly calledHutongs. Nowadays, after years of dramatic revolution and moder-nization, not all alleys retain their original historical form. About 2073alleys, accounting for 13% of all, survive under the historical blockprotection policy and keep their traditional status as Hutongs, these arehighlighted in red in Fig. 1. As a footnote, this map is generated fromthe 2015 road/streets map of Beijing obtained from a local road navi-gation firm, and all of the Hutongs have been simplified into single-linestreets using ArcGIS.

3.1.2. HutongsHutongs are narrow pathways or alleys enclosed by two rows of

traditional siheyuan. In the Yuan Dynasty, the city was divided intoseveral rectangular residential blocks by intersecting main roads. Ineach residential block, narrow streets and Hutongs were interlaced toform the basic skeleton (Fig. 2). The width of a Hutong is 6 steps(traditional Chinese unit of length), about 9.3m in metric units, servingas the corridors for residents’ daily life in that time. The skeleton issmall-scale and has few trees beside the wooden residential buildings toavoid fires. In a nutshell, Hutongs typically represent the fabric ofhistorical streets and are of significant importance to Beijing. From theperspective of historical value, there is an urgency for a technical re-cognition and interpretation of the characteristics of traditional streetforms. The restoration, regeneration and transformation of Hutongspaces are key to historical protection in Beijing, and need continuousscientific observation. While this paper emphasizes mostly the metho-dology, the completeness and pertinence of observations of Hutongs arerelatively preliminary, and waiting for further research.

3.1.3. Street view pictures from TencentThis study uses the SVP dataset from Tencent to investigate the

visual quality of Hutongs. The Tencent Map (http://map.qq.com) wasselected to make use of its ‘Time Machine’ function. A user can find the‘Time Machine’ button once entering into the Tencent SVP browser,which is a database of all past images between 2012 and 2016. As a


4

http://pulse.media.mit.edu

http://pulse.media.mit.edu

http://map.qq.com

departure from previous studies, data is captured directly from theTencent website with open source navigation script processing,building a simulation browser rather than requesting through theTencent SVP API, because historical SVPs could not be captured directlyby the API.

Data acquisition is divided into three steps. First of all, we generateinvestigation points of Hutongs from the above single-line map at in-tervals of 50m, each point with latitude and longitude parameters tofacilitate the following scrawling process. Second, ‘Panos’ was obtained,the Tencent view identification, from geographical coordinates (Long,Liu, & Tang, 2016). Then with parameters including size, ID, pitch(pitch angle), heading (yaw angle) (In our this research,: size:1280 * 720; pitch: 0; heading: 0, 90, 180, 270 for four directional SVPs),and the request Key, we could get an URL to open the street viewwebsite pages.

The following is an URL example (for Pano sample in this link:10011022120723095812200):

http://apis.map.qq.com/ws/streetview/v1/image?size=600x480&pano=10011022120723095812200&pitch=0&heading=0&key=OB4BZ-D4W3U-B7VVO-4PJWW-6TKDJ-WPB77.

Afterward, with the support of NodeJS, CasperJS and SlimerJS, wesimulate the artificial crawling process from is simulated through

entering into a Tencent SVP browser, clicking the Time machine buttonand taking an SVP screenshot. Through the above procedures, we snapsof multi-year SVPs of four directions, namely north, west, east andsouth, are obtained, the stitching of which forms a panoramic view.Finally, 51,356 images for 1892 locations are captured (accessed onMay 18, 2016, some locations have multi-month data in one year).

3.1.4. Other dataGIS data of buildings are also used, each of which has a footprint in

polygons and including building height. There are 213,688 buildings inthe old city in total. The buildings are used for calculating streetscapeparameters to feed the vQoSS evaluation model.

3.2. Methodology

Fuelled by the above questions, a method framework is built con-taining three main parts, which is shown in Fig. 3.

(1) Automatic evaluation of physical vQoSS is done by selecting thetypical physical feature variables, exploring an auto-calculatedapproach with the support of 3-dimensional composition calcula-tion, including machine-learning image segmentation technology,and 2-dimensional GIS analysis. Greenery, openness, enclosures,street wall continuity and cross-sectional proportion are taken asthe representative elements; these five measurable aspects (Fig. 4)are selected according to the literature review and their capacity asmeasurable aspects, to demonstrate the methodology. After this, thesingle-dimension indicator results are combined into one index torepresent the physical vQoSS.

(2) Subjective evaluation of perceived vQoSS is the score of willingnessof street users to remain in the space, sampled from urban designersand borrowing widely accepted evaluation criteria from existingliterature. This study uses enclosure, human-scale, transparency,tidiness and imageability as its standards and compares the physicalvQoSS of streets with the perceived vQoSS of streets as a verifica-tion process. The perceived vQoSS (willingness to spend time in thespace) is measured by a rating (score from 1 to 5) and taken as thebenchmark to check the gap between automatic calculation of re-sults and traditional subjective evaluation results.

(3) The identification of temporal variation of vQoSS of streets by de-tecting physical changes of streets follows. This study sectionsstreets into four components, the façade, pedestrian space, motorroad space, and the commercial space, or wall. The details of eachpart of the methodologies are as follows.

3.2.1. Physical vQoSS evaluation using SegNet and ArcGISSegNet and ArcGIS were chosen for the physical vQoSS measure-

ment. According to the literature, physical vQoSS should consider en-closure, greenery, openness, scale, imageability, transparency and soon, but until now, the subtle factors such as tidiness, imageability, andtransparency have been difficult to achieve through an auto-calculated

Table 2New methods for variation detection.

Category Sample Research method

A. Perceived variation detection basedon algorithm

Naik et al. (2015) Based on the measurement method of Streetscore published in Naik et al. (2014), calculationof Urban Change Coefficient of streets from 2007 to 2014.

B. Automatic physical variationdetection

Taneja, Ballan, and Pollefeys (2011) Using low-resolution images to detect geometric changes for urban environment withalgorithm.

Sakurada and Okatani (2015) Detection of variation of streetscape from street image pair using Convolutional NeuralNetworks (CNNs), features and super-pixel segmentation

Alcantarilla, Stent, Ros, Arroyo, andGherardi (2016)

Detection of change in street video clip by deep CNNs.

C. Physical variation detection byscoring

Tang et al. (2016) Division of the street space into four parts according to sections and identification oftemporal variation by scoring.

Fig. 1. Map of Second Ring Road and Hutongs in Beijing.


5

http://apis.map.qq.com/ws/streetview/v1/image?size=600x480%26pano=10011022120723095812200%26pitch=0%26heading=0%26key=OB4BZ-D4W3U-B7VVO-4PJWW-6TKDJ-WPB77



method. So rather than construct a concrete and all-inclusive indexsystem, more effort is taken to construct an integrated method.Therefore, five important and measurable indicators are selected first:greenery, openness, and enclosure with the SegNet decoder (for thebasic Algorithm principle, see: Kendall et al., 2015; Badrinarayananet al., 2015), and then the cross-sectional proportion and the street wallcontinuity are measured with the newly emerging GIS approach pro-vided by Harvey (2014). A lot of semantic pixel-width image segmen-tation methods based on convolutional networks have emerged re-cently, such as YOLO, ImageNet, SegNet, DeepLab and so on. SegNet ishighly accurate and easily accessible. Fig. 5 is a segmentation demo bySegNet from their website (http://mi.eng.cam.ac.uk/projects/segnet/).

SegNet is a fully trainable approach for joint feature learning andimage mapping to pixel-wise semantic labels, which turns out to be thefirst deep learning method to attach low resolution features with se-mantic labels (Kendall et al., 2015; Badrinarayanan et al., 2015). Theconvolutional neural network architecture of SegNet uses a set of en-coders and decoders followed by a pixel-wise classifier as a coretrainable engine (Badrinarayanan, et al., 2016), where the object re-cognition is achieved by large-sample labeling and training instead ofcolor extraction. For indicators like greenery, openness, and enclosure,which are features that depend on the street composition ratio, SegNetcould simplify the calculation approach and automate the interpreta-tion of the streetscape. For example, for the recognition of “tree”, thetraining label combines form and other features of trees, reducing thevariability of seasons on the mapping. The following test samples ofSegNet segmentation prove that tree form was ascertained correctly,even in leafless seasons and dusk scenes (Fig. 6). After testing a varietyof outdoor RGB images, SegNet has been considered competitive in its

measurement of the proportion of street elements, especially forbuildings, column-poles, trees and the sky, and it has been used as abenchmark for the following improved segmentation method (Chenet al., 2016; Kundu, Vineet, & Koltun, 2016).

All of the most recent year’s (2016) SVPs are interpreted into colorgroups using the SegNet decoder. Twelve kinds of physical componentsare obtained, namely sky, buildings, columns-poles, road markings,roads, pavement, trees, sign symbols, fences, vehicles, pedestrians, andbikes. We summarize percentages of each component in the four-di-rectional (North, South, East, West) dataset (7568 images for 1892 lo-cations) are summarized and the mean value for each location is cal-culated to represent the average condition of that location. Then thepercentage of tree as greenery (%), sky ratio as openness (%), and thetotal sum of buildings, columns and trees as enclosure (%) are used. Thetotal calculation is based on the resulting proportions calculated bySegNet.

In addition, methods are borrowed from Harvey (2014) using theArcGIS geo-processing function to calculate cross-sectional proportionand street wall continuity. The method is a creative way to combine GISwith SVP data, automatically obtaining these two key physical in-dicators. Specifically, cross-sectional proportion is the ratio of averageheight to street width, which could be calculated directly for both sides.The height is the average height of the buildings. When there are onlywalls instead of buildings, it calculates with the height of the wallsalongside, depending on their actual objective existence. While for thestreet wall continuity, the street is firstly broken into segments, 40single-sided centerline buffers are then created at 1m intervals, calcu-lating the ratio of the area of each buffer and the area without thebuildings. The street edge is selected out with SQL when the ratio

Fig. 2. Peking and its Environs 1912 Map Source: Madrolle's Guide Books, http://www.lib.utexas.edu/maps/historical/history_china.html.


6

http://mi.eng.cam.ac.uk/projects/segnet/

http://www.lib.utexas.edu/maps/historical/history_china.html

reaches its peak. Continuity is the proportion of the length of the edgeto the total length of the street segment. After this follows the calcu-lation of the mean values of cross-sectional proportion and continuity ofboth sides as the value for that segment.

Thus, several indicators (a temporary five in this research) havebeen formed to comprehensively evaluate the physical vQoSS, the va-lues of which ai are normalized to the values zi based on the deviationstandardization Eq. (1):

=z a aa a

min( )max( ) min( )i

i i

i i (1)

i= greenery, openness, enclosure, wall continuity, cross sectionalproportion and so on.

Following this, these five normalization values are summarized intoone Index bpquality according to Eq. (2), below, which represents theoverall physical vQoSS. Since the main purpose of this paper is toprovide a methodology for quantitative measurement of visual qualityof streetscapes in the empirical study of Hutongs the indicators aresimplified into five and the coefficients of each are equally distributedfor the convenience of discussion. Accurate coefficients still requirefurther quantitative research based on large-sample surveying andtheoretical literature reviews.

=b apqualityi

i i(2)

i are the weighting coefficients, awaiting further detailed researchfor improvement.

3.2.2. Evaluation of perceived vQoSS (Stay willingness scoring)The willingness to stay (in a space) is scored by urban designers to

measure the overall perceived vQoSS directly, borrowing criteria fromEwing and Clemente (2013). The five aspects represent the key aspectsof perceived vQoSS, and the criteria have already been verified tosuccessfully measure the previously unmeasured street quality. In orderto clearly make comparison with the physical vQoSS result, they aresummarized into one index, called the ‘Stay willingness score’.

According to Ewing and Clemente (2013), imageability refers to thedistinguishable characteristics of specific space; Lynch (1960) believesthat space with good imageability is well shaped with special compo-nents for users to recognize, such as landmarks, walls, doorknobs, ordomes, by which the street space gains a unique style; for the score ofimageability, the raters value whether the space generates a feeling ofrecognizability, symbolism, indication of, or a sense of place or per-ception of locality. The more features above, the higher the score of

Fig. 3. The framework of the study, where ‘Stay willingness score’ is an individual’s willingness to use or remain in a space.


7

Fig. 4. Five selected aspects for physical vQoSS calculation in this paper. (A) Street wall continuity (B) Cross-sectional proportion (C) Greenery (D) Openness (E)Enclosure.

Fig. 5. A segmentation demo by SegNet.


8

imageability. The meaning of enclosure has been explained in Section2.1. Human-scale, which describes the scale and proportion of thephysical street space, could affect psychological feelings throughchanges in architectural details, pavement forms, and greenery; trans-parency refers to the extent to which activities in the public street spacecould be visually accessed through transparent facades by the pedes-trian, while tidiness is easy to understand, measuring the level ofcleanliness or pollution and garbage.

Our four raters are all designers with a background in urban designeducation, and they were invited to comprehensively measure thevQoSS from the above five aspects, giving a score from 1 to 5. To avoidresults from being affected by seasonal weather and subjective bias,they firstly scored together to have a consensus of the criteria, and thenscored separately, which lasted nine days in total. Fig. 7 shows thesample SVPs for different Stay willingness scores.

3.2.3. Variation recognitionVariation recognition aims to reflect the appearance and changes in

visual quality through variation of the street elements indirectly. Usingcurrent technology, the variation recognition is typically difficult toautomate. Although multi-year SVPs for one location are initially in-tended to record the change for single sites, the recording angle couldnot avoid subtle deviation from year to year. As such, the artificialidentification is still the most reliable method, which, however, is yetbiased due to subjective judging. Therefore, this research aims at arti-ficial recognition of variation by dividing the whole street section intofour parts (Fig. 8) to minimize the resulting errors. From these fourparts, the changing objects are identified in each area, then judged onwhether those variations are effective for vQoSS improvement or not.As it shows in the evaluation decision tree (Fig. 9), each part has two orthree elements to be checked (score 0 or 1) and an effective or non-effective indicator. If the variations have positive influence on vQoSS,

then the score is 2, otherwise the score is 1 (No change is 0). Together,all these 11 sub-indicators make up a framework for observation. SVPsin Fig. 10 are the scoring samples for variation recognition.

4. Results

4.1. Visual quality of Hutongs: Physical features

Fig 11 shows the final five-aspect physical vQoSS index result andFig. 12 shows the five physical features’ results separately. The averagephysical quality index score is 1.39. The results are verified by com-paring high scoring areas with site visits: it was found that the indexcould tell the physical form, but perceived vQoSS of (5) and (8) are low,though they are equipped with good form due to tidiness conditions.This means the index is accurate for physical form, but not for reflectingthe comprehensive visual quality. It can be concluded that Hutongs onthe north side of Chang’an Road have better skeletons than those on thesouth, especially Hutongs in the historical protection areas; those nearthe traditional imperial gardens (like Houhai, Shishahai, Xihai, andQianhai), palaces, temples, traditional communities (Fangjia Hutong,Dongsi Hutong Area), and regeneration areas (Dashilan Area, Nanluo-guxiang) received the highest score. It is not difficult to observe thatsome regeneration projects take full consideration of the historicalforms (Fig. 13 (a)), keeping the original architectural features, whilesome partially change in spatial scale due to the need for moderntransportation (Fig. 13(b)). Hutongs close to the main roads have ahigher degree of openness and a relatively low degree of enclosure,indicating dramatic changes in the skeleton and widening processescatering to the demands of modern vehicles. In addition, in these Hu-tongs, the percentages of road is much higher than that of others. Thecross-sectional proportion is generally lower than 0.5, while the con-tinuity is around 0.7. Joint distributions of these two indicators show

Fig. 6. test samples of SegNet source: Badrinarayanan, et al. (2016).

Fig. 7. SVP samples for Stay willingness score ranging from 1 to 5.


9

that less than 50% of Hutongs keep their compact skeleton, while othersare replaced by porous forms.

In order to further portray the character of Hutongs and reflect theirvisual quality, an additional comparison is provided, calculating thephysical component ratios of streets in Shanghai’s Hengfu historicalarea, Beijing, and the Shanghai central area using the same method.The Hengfu historical area was previously the Shanghai FrenchConcession, which retains western style streets and architecture. (Data

was lacking for the other two aspects, otherwise all five aspects wouldhave been compared together in the national context.) Fig. 14 showsthe results for these four regions. From the comparison, it was foundthat most Hutongs are lacking visual greenery; the average percentageof greenery (19.8%) is similar to the central area of Beijing, but lowerthan that of the other three areas. Enclosures were much lower at46.1%, because most Hutongs are 1-floor residential buildings with lesstree shade and occlusions. These numeric results in the form illustrate a

Fig. 8. Section of the street with commercial part along the Hutong (left) and without the commercial part, but with wall along the Hutong (right) Pink: Commercialspace frontage; Orange: Street façade. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 9. Criteria for variation recognition.


10

Fig. 10. Samples for four part variation recognition.


11

mid-level quality of Hutongs, which were lower than expectation.

4.2. Visual quality of Hutongs: Perceived stay willingness

Fig. 15 shows the overall Stay willingness scores of Hutongs. Theaverage score is 2.59, where only 1.2% are rated 5, and 8.8% are rated1. Most Hutongs are rated 2 or 3, accounting for 39.3% and 39.8%respectively. The result shows us the subjective understanding of

Hutongs, that the overall perceived vQoSS is not one of satisfaction,except in some few areas (Shichahai, Nanluoguxiang) boasting improvedHutongs and roads (Ganmian Hutong, Lumicang Hutong, Xigong Road,Middle Jingji Road), which received higher ratings.

High score area in Fig. 14: (1) Northeast to the Shishahai, (2) MaoerHutong and Nanluoguxiang, (3) Xisi Area, (4) Ganmian Hutong andLumicang Hutong, (5) Middle Jingji Road, (6) Xigong West and EastRoad

Fig. 10. (continued)


12

These results enabled the verification of the result of previousphysical features calculated by automatic measurement. Taking theperceived vQoSS (the Stay willingness score) rated by educated de-signers as the benchmark, there is a discrepancy between the D-value ofnormalized perceived and physical vQoSS. Areas with inconsistent re-sults are identified, which are shown as blue in Fig. 16, by the StandardDeviation classification, vQoSS of these locations is further verified bySVPs and site investigations. Results show that the main contributingfactor to the difference between the approaches is the level of ignoranceof aspects related to the degree of decline and human activity, such asthe level of dilapidation on the street, tidiness, level of chaos, and vi-tality in the SVP auto-calculation process; while these factors are im-portant indicators for subjective evaluation, they are missing from au-tomated evaluation, leaving a gap for further research.

4.3. Visual quality of Hutongs: Temporal variation

Table 3 gives detailed statistical results for variation identification.Generally speaking, there has been very little variation in the elementsover the past 3–4 years, mostly less than 10%. The total amount ofvariation varies from street section to section. Most changes take placein street façade and commercial real estate, accounting for 10% (StreetFaçade, Façade color), 9% (Street Façade, Façade material or decora-tion), 16% (Commercial, Signboards change) and 9% (Commercial,Store façade transparency, decoration change) respectively (seeTable 3, below), possibly resulting from residential beautification and

rapid store retail turnovers and replacements. 8% of the walls along thestreet changed transparency, 7% had changes with greenery and facil-ities construction, which could have promoted communication with thestreet. Only 284 Hutongs experienced variation regarding pedestrians,and 165 had road changes, accounting for around 5% variation. It iseasy to conclude that vQoSS changes slowly and infrequently.

In addition, most of the variations are not satisfactory. The mosteffective improvements occur in the commercial section, driven by themarket (1.7%), and the street façade (2.5%). Other sections account forless than 1.5% of effective improvements in total (Table 4). The renewalprocess for Hutongs seems to deliver unfavorable outputs, and it seldomreflects the concepts and principles of smart design, so more efforts areneeded for urban regeneration with respect to the surrounding histor-ical context. In other words, the improvement of the visual quality ofHutongs has been and will be a lengthy process.

5. Conclusion and discussion

5.1. Concluding remarks

This study proposed an evaluation framework for vQoSS, includingphysical and perceived vQoSS and assessment of its temporal variation.The first was calculated automatically with the support of SegNet, whilethe latter two were achieved quantitatively through a rating systembased on systematic indicators. The realization of this combinedmethod exploits new applications of Street View Picture and new

Fig. 11. Physical vQoSS Index of Hutongs. High scoring areas: (1) Eastern to Xihai, (2) North to Xinjiekou Street, (3) South to Gungwangfu, (4) Nanluoguxiang, (5)Fangjia Hutong, (6) North to Xisi Station, (7) Dongsi Hutong Area, (8) Xiting and Dongting Area, (9) Xixinglong Street, (10) Dashilan Area.


13

emerging image segmentation techniques. With perceived vQoSS as thebenchmark, the paper concludes the strengths and weaknesses of theautomatic calculated method of physical vQoSS.

An empirical study of Hutongs shows that, although the Old City ofBeijing has gained a reputation for its cultural, historical and archi-tectural design values, the general visual quality fails to satisfy thecurrent needs, remaining to be ameliorated. Physical feature analysisindicates Hutongs in the historical protection areas have better physicalform, while others receive lower scores. However, joint analysis ofstreet continuity and cross-sectional proportion show that less than halfof the Hutongs have maintained their previous forms and structure

while others are reconstructed in a relatively modern approach, re-placed by porous forms. Many traditional styles and features in theinner city are challenged by modernization. The inner city is still a locusattracting developer attention. The process of improvement is an on-going balance between the traditional and modern. Nevertheless, betterskeletons do not mean better visual quality, –the degree of decline andhuman activity should be taken into consideration for the overallquality.

The variation results indicate that visual quality change is un-obvious and slow; less than 10% of Hutongs have changing elements,while no more than 2.5% are effective in the past 3–4 years, indicating

Fig. 12. Physical features of Hutongs.

Fig. 13. (a) left, Dashilan West street (regeneration) (b) right, Nanchizi Street.


14

the difficulties in renovation and regeneration of historical areas in-volved in chaotic property rights and interest-balancing issues.According to the site survey conducted later, the inner city is mixedwith a variety of functions and property rights, varying from hostels,regenerated commercial streets, and low rent housing to residentialhouses. Therefore, to give Hutongs a new life and obtain a high-qualityenvironment, public participation and environmental awareness arequite necessary. In addition, despite the less distinctive trend in en-vironmental improvement, they show no sign of decline.

5.2. Potential applications

This study is an initial attempt to construct a quantitative researchmethodology for vQoSS. After testing, this methodology shows its fea-sibility and convenience for preliminary investigation of urban design,suggesting potential application as an augmented tool for designers toeliminate subjective judgment bias and criticism, and for enabling amore rational design. Considering the convenience and accessibility ofdata acquisition, it could be applied beforehand in evaluation, designelements extraction, and analysis of design implementation, so as tofacilitate a feedback process, for example, as an assist-tool for the

Fig. 14. Comparative analysis of greenery, openness, and enclosures between Hutongs in Beijing, the Shanghai Hengfu historical area, and the Beijing and Shanghaicentral areas. Footnote: the central areas in the two comparable largest cities, Beijing and Shanghai in China, were selected because their central areas remainrelatively traditional after long-term urban development, of which the spatial scale is more comfortable and has been better maintained. The regions outside thecentral area are mostly newly developed high-rise blocks with wide roads. Therefore, it makes no sense to compare Hutongs to them. Secondly, to be specific, theShanghai Hengfu historical area was picked as a reference because this area used to be a French Concession with a compact scale, pleasant greenery and goodimageability.

Fig. 15. Stay willingness scores of Hutongs.


15

assessment of master plans and comprehensive urban design. Designerscould discover the key aspects and precise positions of street space forrefinement, so that they could optimize the design and realize the op-timal combination of science, practicality and aesthetics. Segmentationand objective physical component analysis aim to contribute empiricalvalues for smart design guideline making in different function areas,such as residential, historical, business areas, and so on. In addition, thevisual quality evaluation process also shows its potential use in stimu-lating participative planning. With a variety of evaluation platforms,street space quality can be rated directly by the residents, promotingusers to express their expectations for the environment and forming apositive interaction between the citizens and the designers, truly en-abling “streets for the people”.

5.3. Academic contributions

Beside the possibility for applying this method in urban designprojects, this study also contributes a method for measuring the builtenvironment as follows. First, the Street View Picture is proven to be alarge-scale database, matching the need for visual quality data of thebuilt environment. With the continuing development of ImageProcessing Technology, SVP may bring further findings in this field.Visual quality is one of the most important dimensions of the builtenvironment. Using micro-scale SVP data, this study achieves auto-calculated physical vQoSS based on GIS and segmentation technology,and identifies the limitations by mutual verification and rectificationwith perceived vQoSS. Based on classical theories, a methodologicalframework is constructed for quantitative measurements of thestreetscape, which is available for wide application. The partially in-consistent result provides clues for the potential for subsequent studiesof detailed coefficients. Second, by extending the scale of the mea-sureable built environment, the pattern of quality could be measured ona large scale with fine-granularity data (Fig. 17) from the previously

Fig. 16. D-value of perceived and physical vQoSS (Normalized by perceived vQoSS).

Table 3Statistical result for variation.

Indicators N Mean Total change

Street Façade facade color (X1) 1852 0.10 X1+X2=352facade material ordecoration (X2)

1852 0.09

Pedestrian Parking(X3) 1891 0.04 X3+X4+X5=284Greenery (X4) 1891 0.07Street furniture increasesor additional optimization(X5)

1891 0.04

Road Lane refinement (X6) 1498 0.06 X6+X7=165Greenery improvement(X7)

1497 0.05

Commercial Signboards change (X8) 845 0.16 X8+X9=211Store facade transparency,decoration change (X9)

842 0.09

Wall Transparency change(X10)

433 0.08 X10+X11=68

Surrounding greenery andfacilities construction(X11)

441 0.07

Table 4Effectiveness of variation.

The Effectiveness ofChange

N Proportion of the effectivechange

Street Façade Non-Effective 7 3.5%Effective 16 2.5%

Pedestrian Non-Effective 82 4.6%Effective 9 0.5%

Road Non-Effective 56 3.9%Effective 10 0.7%

Commercial Non-Effective 47 5.6%Effective 14 1.7%

Wall Non-Effective 19 4.4%Effective 6 1.4%


16

restricted community-scale to the current scale (including street, com-munity and city) which comprehensively demonstrates the physical andsocial attributes of street space. Thus, the visual images of a city couldbe reflected through stitching streetscape images rather than onlythrough qualitative description, while the large-scale calculation resultalso lays a foundation for potential investigation into the relation be-tween the micro-built environment to macro-scale socioeconomic at-tributes in the city region. Third, this study provides an inter-disciplinary perspective on built environment studies. Computertechnology is foreseen to bring a qualitative leap through inter-disciplinary integration of urban design and computing technology.Semantic image segmentation methods, like SegNet’s, applied in thisresearch, demonstrate a new approach for visual deconstruction, whichmay hopefully promote more automatic methods for smart design in thefuture.

5.4. Potential limitations and future research

While appreciating the merits of this study, there are still severallimitations deserving further research in the near future. First, theconcept of visual quality is human-dependent (subjective), consideringits nature, based in sensation and perception. The scoring index forvisual quality of street space may vary among different people groups,and future studies are expected to address this issue. Firstly, despiteplenty of relevant theories and existing methods reviewed at the be-ginning of the article to ensure rationality and effectiveness, the se-lection of integrated indicators still suffers from the limitations ofquantitative techniques. Second, the determination of coefficients ( i)are still waiting for further comprehensive discussion, for which thestreetscape typology based on function, location and hierarchy ofstreets are fundamental. Streetscape with more recreational function issupposed to give more weight to aesthetic indicators while for mainstreets focusing on mobility is on the opposite; In addition, the per-ceived quality of raters/users‘ judgments may be significantly affectedby their expectation for vQoSS, therefore, the typology classificationbefore data collection and rating is worth for cautious discussion. Andfor variation recognition, long-term field tracking surveys and corre-sponding time-serial SVP data are expected to be collected simulta-neously, as a mutual verification of the computer-augment image re-cognition by traditional methods. Third, it was found that the physicalvQoSS could only partially explain the overall vQoSS due to the ig-norance of the degree of dilapidation, tidiness, chaos, and human vi-tality in this study, and the contribution of each indicator was therebymade unclear, requiring one-by-one detection with large sample data;special attention is required to bridge this gap. The best method forovercoming the limitation of perceived senses in the future, using in-telligent technology to integrate the visual and non-visual factors, andproviding an accurate guideline for city design implementation, is a

topic needy of continuous discussion.

Acknowledgements

This work was supported by the National Water Pollution Controland Treatment Science and Technology Major Project of China (GrantNo. 2017ZX07103-002), the National Nature Science Foundation ofChina (Grant No.51778319), and Tsinghua University (School ofArchitecture) CIFI Group Joint Research Center for SustainableCommunity (Grant No.R201).

References

Aiello, L. M., Schifanella, R., Quercia, D., & Aletta, F. (2016). Chatty Maps: Constructingsound maps of urban areas from social media data. Royal Society Open Science, 3,150690. https://doi.org/10.1098/rsos.150690.

Alcantarilla, P.F., Stent, S., Ros, G., Arroyo, R., & Gherardi, R., (2016). Streetview changedetection with deconvolutional networks. In: Robotics: Science and systems con-ference (RSS).

Aletta, F., Lepore, F., Kostara-Konstantinou, E., Kang, J., & Astolfi, A. (2016). An ex-perimental study on the influence of soundscapes on people's behaviour in an openpublic space. Applied Sciences, 6(10), 276.

Alexander, C., Ishikawa, S., Silverstein, M., Jacobson, M., Fiksdahl-King, I., & Angel, S.(1977). A pattern language: Towns, buildings, construction. New York: Oxford UniversityPress.

Appleyard, D., & Lintell, M. (1972). The environmental quality of city streets: The re-sidents' viewpoint. Journal of the American Institute of Planners, 38(2), 84–101.

Arnold, H. F. (1980). Trees in urban design. Trees in urban design.Aspinall, P., Mavros, P., Coyne, R., & Roe, J. (2013). The urban brain: Analysing outdoor

physical activity with mobile. EEG. British Journal of Sports Medicine bjsports-2012.Bader, M. D., Mooney, S. J., Lee, Y. J., Sheehan, D., Neckerman, K. M., Rundle, A. G., &

Teitler, J. O. (2015). Development and deployment of the Computer AssistedNeighborhood Visual Assessment System (CANVAS) to measure health-relatedneighborhood conditions. Health & Place, 31, 163–172.

Badrinarayanan, V., Handa, A., & Cipolla, R. (2015). Segnet: A deep convolutional en-coder-decoder architecture for robust semantic pixel-wise labelling. arXiv preprintarXiv:1505.07293.

Brownson, R. C., Hoehner, C. M., Day, K., Forsyth, A., & Sallis, J. F. (2009). Measuring thebuilt environment for physical activity: State of the science. American journal ofpreventive medicine, 36(4), S99–S123.

Bruce, V., Green, P. R., & Georgeson, M. A. (2003). Visual perception: Physiology, psy-chology, & ecology. Psychology Press.

Carozza, L., Tingdahl, D., Bosché, F., & Gool, L. (2014). Markerless vision-based aug-mented reality for urban planning. Computer-Aided Civil and Infrastructure Engineering,29(1), 2–17.

Cassidy, T. (1997). Environmental psychology. London, UK: Psychology Press.Chen, L. C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2016). Deeplab:

Semantic image segmentation with deep convolutional nets, atrous convolution, andfully connected crfs. arXiv preprint arXiv:1606.00915.

Curtis, J. W., Curtis, A., Mapes, J., Szell, A. B., & Cinderich, A. (2013). Using google streetview for systematic observation of the built environment: Analysis of spatio-temporalinstability of imagery dates. International Journal of Health Geographics, 12(1), 53.

Dover, V., & Massengale, J. (2013). Street design: The secret to great cities and towns. JohnWiley & Sons.

Dupont, L., Antrop, M., & Van Eetvelde, V. (2014). Eye-tracking analysis in landscapeperception research: Influence of photograph properties and landscape character-istics. Landscape Research, 39(4), 417–432.

Easteal, M., Bannister, S., Kang, J., Aletta, F., Lavia, L., Witchel, H.J. Urban soundplanning in Brighton and Hove. In Proceedings of the Forum Acusticum, Krakow,Poland, 7–12 September 2014.

Ewing, R., & Clemente, O. (2013). Measuring urban design: Metrics for livable places.Washington, DC: Island Press.

Ewing, R., & Handy, S. (2009). Measuring the unmeasurable: Urban design qualities re-lated to walkability. Journal of Urban Design, 14(1), 65–84.

Fruin, J. J. (1971). Pedestrian planning and design. New York: Metropolitan Association ofUrban Designers and Environmental Planners.

Fyfe, N. (Ed.). (2006). Images of the street: Planning, identity and control in public space.Routledge.

Gehl, J. (2013). Cities for people. Washington, DC: Island press.Gehl, J., & Gemzøe, L. (1996). Public spaces-public life. Copenhagen: Danish architectural

press.Gehl, J., & Svarre, B. (2013). How to study public life. Washington, DC: Island Press.Grimmond, C. S. B., Potter, S. K., Zutter, H. N., & Souch, C. (2001). Rapid methods to

estimate sky view factors applied to urban areas. International Journal of Climatology,21(7), 903–913.

Harvey, C., Aultman-Hall, L., Hurley, S. E., & Troy, A. (2015). Effects of skeletalstreetscape design on perceived safety. Landscape and Urban Planning, 142, 18–28.

Harvey, C. W. (2014). Measuring Streetscape Design for Livability Using Spatial Data andMethods.

Heath, G. W., Brownson, R. C., Kruger, J., Miles, R., Powell, K. E., & Ramsey, L. T. (2006).The effectiveness of urban design and land use and transport policies and practices toincrease physical activity: A systematic review. Journal of Physical Activity and Health,

Fig. 17. Measurable scale expansion achieved by combining subjective andobjective evaluation (in grey).


17

https://doi.org/10.1098/rsos.150690

http://refhub.elsevier.com/S0169-2046(18)31011-9/h0010



















































3(s1), S55–S76.Henshaw, V. (2014). Urban smellscapes – Understanding and designing city smell environ-

ments. Abingdon-on-Thames, UK: Routledge.Jacobs, J. (1961). The death and life of great American cities. New York: Random House.Jiang, B., Chang, C. Y., & Sullivan, W. C. (2014). A dose of nature: Tree cover, stress

reduction, and gender differences. Landscape and Urban Planning, 132, 26–36.Jiang, B., Larsen, L., Deal, B., & Sullivan, W. C. (2015). A dose–response curve describing

the relationship between tree cover density and landscape preference. Landscape andUrban Planning, 139, 16–25.

Johansson, M., Sternudd, C., & Ferreira, I. (2015). The Walkshop: A tool to integrateresearch on human aspects of sustainable urban design in teaching. Högre utbildning,5(3), 145–157.

Kaplan, S. (1995). The restorative benefits of nature: Toward an integrative framework.Journal of Environmental Psychology, 15, 169–182.

Kaplan, R., & Kaplan, S. (1989). The Experience of nature: A psychological perspective.Cambridge: Cambridge University Press.

Kendal, D., Williams, N. S., & Williams, K. J. (2012). Drivers of diversity and tree cover ingardens, parks and streetscapes in an Australian city. Urban Forestry & UrbanGreening, 11(3), 257–265.

Kendall, A., Badrinarayanan, V., & Cipolla, R. (2015). Bayesian segnet: Model uncertaintyin deep convolutional encoder-decoder architectures for scene understanding. arXivpreprint arXiv:1511.02680.

Kundu, A., Vineet, V., & Koltun, V. (2016). Feature space optimization for semantic videosegmentation. In Proceedings of the IEEE conference on computer vision and patternrecognition (pp. 3168–3175).

Li, X., Ratti, C., & Seiferling, I. (2017). Mapping urban landscapes along streets usinggoogle street view. International cartographic conference (pp. 341–356). Cham:Springer.

Li, X., Zhang, C., & Li, W. (2015). Does the visibility of greenery increase perceived safetyin urban areas? Evidence from the Place Pulse 1.0 Dataset. ISPRS International Journalof Geo-Information, 4(3), 1166–1183.

Li, X., Zhang, C., Li, W., Ricard, R., Meng, Q., & Zhang, W. (2015). Assessing street-levelurban greenery using Google Street View and a modified green view index. UrbanForestry & Urban Greening, 14(3), 675–685.

Lin, T. P., Tsai, K. T., Hwang, R. L., & Matzarakis, A. (2012). Quantification of the effect ofthermal indices and sky view factor on park attendance. Landscape and UrbanPlanning, 107(2), 137–146.

Long, Y., Liu, L., & Tang, J. (2016). How green are streets? An analysis for 245 majorChinese cities using Tencent street view. One PLoS under review.

Luigi, M., Massimiliano, M., Aniello, P., Gennaro, R., & Virginia, P. R. (2015). On theValidity of Immersive Virtual Reality as tool for multisensory evaluation of urbanspaces. Energy Procedia, 78, 471–476.

Lynch, K. (1960). The image of the city. Cambridge: The MIT Press.Lynch, K. (1984). Good city form. Cambridge: The MIT press.McGinn, A. P., Evenson, K. R., Herring, A. H., Huston, S. L., & Rodriguez, D. A. (2007).

Exploring associations between physical activity and perceived and objective mea-sures of the built environment. Journal of Urban Health, 84(2), 162–184.

Middleton, J. (2016). The socialities of everyday urban walking and the ‘right to the city’.Urban Studies, 0042098016649325.

Montgomery, C. (2013). Happy city: Transforming our lives through urban design.Macmillan.

Mounir, G. M., Attia, M. S., & Tayel, K. Y. (2016). Street children in Alexandria: Profileand psychological disorders. Journal of High Institute of Public Health, 37(1), 56–77.

Naik, N., Kominers, S. D., Raskar, R., Glaeser, E. L., & Hidalgo, C. A. (2015). Do peopleshape cities, or do cities shape people? The co-evolution of physical, social, and economicchange in five major US cities (No. w21620). National Bureau of Economic Research.

Naik, N., Philipoom, J., Raskar, R., & Hidalgo, C. (2014). Streetscore–predicting theperceived safety of one million streetscapes. 2014 IEEE conference on computer visionand pattern recognition workshops (pp. 793–799). IEEE.

Nasar, J. L. (1989). Perception, cognition, and evaluation of urban places. In Public places andspaces. Boston, MA: Springer31–56.

Owens, P. M. (1993). Neighborhood form and pedestrian life: Taking a closer look.Landscape and Urban Planning, 26(1–4), 115–135.

Pheasant, R. J., Horoshenkov, K., Watts, G., & Barret, B. T. (2008). The acoustic and visualfactors influencing the construction of tranquil space in urban and rural environ-ments tranquil spaces-quiet places? Journal of the Acoustical Society of America,123(3), 1446–1457.

Rapoport, A. (2016). Human aspects of urban form: Towards a man—environment approachto urban form and design. Elsevier.

Rundle, A. G., Bader, M. D., Richards, C. A., Neckerman, K. M., & Teitler, J. O. (2011).Using Google Street View to audit neighborhood environments. American Journal ofPreventive Medicine, 40(1), 94–100.

Saelens, B. E., Sallis, J. F., & Frank, L. D. (2003). Environmental correlates of walking andcycling: Findings from the transportation, urban design, and planning literatures.Annals of Behavioral Medicine, 25(2), 80–91.

Sakurada, K., & Okatani, T. (2015). Change detection from a street image pair using CNNfeatures and superpixel segmentation. In BMVC (pp. 61-1).

Salesses, P., Schechtner, K., & Hidalgo, C. A. (2013). The collaborative image of the city:Mapping the inequality of urban perception. PLoS ONE, 8(7), e68400.

Sallis, J. F., Johnson, M. F., Calfas, K. J., Caparosa, S., & Nichols, J. F. (1997). Assessingperceived physical environmental variables that may influence physical activity.Research Quarterly for Exercise and Sport, 68(4), 345–351.

Shen, Q., Zeng, W., Ye, Y., Arisona, S. M., Schubiger, S., Burkhard, R., & Qu, H. (2018).StreetVizor: Visual exploration of human-scale urban forms based on street views.IEEE Transactions on Visualization and Computer Graphics, 24(1), 1004–1013.

Slater, K. (1985). Human comfort, Vol. 1. Springfield, Ill., USA: CC Thomas.Taneja, A., Ballan, L., & Pollefeys, M. (2011). Image based detection of geometric changes

in urban environments. Computer Vision (ICCV), 2011 IEEE International Conference on(pp. 2336–2343). IEEE.

Tang, J., & Long, Y. (2017). Evaluation for Quality of Street Space in center area of mega-city: A comparative analysis of Beijing and Shanghai. Planners, 2017(2), 68–73 (InChinese).

Tang, J., Ma, Y., Zhai, W., & Long, Y. (2016). Measuring quality of street space, itstemporal variation and impact factors: An analysis based on massive street viewpictures. New Architecture, 2016(5), 130–135 (In Chinese).

Thompson, C. W. (2002). Urban open space in the 21st century. Landscape and UrbanPlanning, 60, 59–72.

Unger, J. (2008). Connection between urban heat island and sky view factor approxi-mated by a software tool on a 3D urban database. International Journal of Environmentand Pollution, 36(1–3), 59–80.

Wang, X., Kim, M. J., Love, P. E., & Kang, S.-C. (2013). Augmented Reality in built en-vironment: Classification and implications for future research. Automation inConstruction, 32, 1–13.

Ye, Y., & van Nes, A. (2014). Quantitative tools in urban morphology: Combining spacesyntax, spacematrix, and mixed-use index in a GIS framework. Urban Morphology,18(2), 97–118.


18

































































































measuring visual quality of street space and its temporal ...€¦ · “physical vqoss”)...

Documents